This paper aims to provide a roadmap to AI governance. In contrast to the reigning paradigms, we argue that AI governance should not be merely a reactive, punitive, status-quo-defending enterprise, but rather the expression of an expansive, proactive vision for technology—to advance human flourishing. Advancing human flourishing in turn requires democratic/political stability and economic empowerment. Our overarching point is that answering questions of how we should govern this emerging technology is a chance not merely to categorize and manage narrow risk (...) but also to construe the risks and opportunities much more broadly, and to make correspondingly large investments in public goods, personnel, and democracy itself. To lay out this vision, we take four steps. First, we define some central concepts in the field, disambiguating between forms of technological harms and risks. Second, we review normative frameworks governing emerging technology that are currently in use around the globe. Third, we outline an alternative normative framework based in power-sharing liberalism. Fourth, we walk through a series of governance tasks that ought to be accomplished by any policy framework guided by our model of power-sharing liberalism. We follow these with proposals for implementation vehicles. (shrink)

The stakes associated with an algorithmic decision are often said to play a role in determining whether the decision engenders a right to an explanation. More specifically, “high stakes” decisions are often said to engender such a right to explanation whereas “low stakes” or “non-high” stakes decisions do not. While the overall gist of these ideas is clear enough, the details are lacking. In this paper, we aim to provide these details through a detailed investigation of what we will call (...) the “Simple Stakes Thesis”. The Simple Stakes Thesis, as it will turn out, is too simple. For even if the stakes associated with a specific one-off decision are low—and hence does not engender a right to an explanation—such decisions may nevertheless form part of a high stakes pattern or aggregate of decisions. In such cases, we argue, even a low stakes decision may engender a right to explanation. Not only does this show that the right to explanation is more demanding than so far recognized. It also shows that the stakes thesis is significantly harder to apply in practice. (shrink)

The literature on algorithmic fairness has examined exogenous sources of biases such as shortcomings in the data and structural injustices in society. It has also examined internal sources of bias as evidenced by a number of impossibility theorems showing that no algorithm can concurrently satisfy multiple criteria of fairness. This paper contributes to the literature stemming from the impossibility theorems by examining how informational richness affects the accuracy and fairness of predictive algorithms. With the aid of a computer simulation, we (...) show that informational richness is the engine that drives improvements in the performance of a predictive algorithm, in terms of both accuracy and fairness. The centrality of informational richness suggests that classification parity, a popular criterion of algorithmic fairness, should be given relatively little weight. But we caution that the centrality of informational richness should be taken with a grain of salt in light of practical limitations, in particular, the so-called bias-variance trade off. (shrink)

Despite growing interest in automated (or algorithmic) decision-making (ADM), little work has been done to conceptually clarify the term. This article aims to tackle this issue by developing a conceptualization of ADM specifically tailored to organizational contexts. It has two main goals: (1) to meaningfully demarcate ADM from similar, yet distinct algorithm-supported practices; and (2) to draw internal distinctions such that different ADM types can be meaningfully distinguished. The proposed conceptualization builds on three arguments: First, ADM primarily refers to the (...) automation of practical decisions (decisions to _φ_) as opposed to cognitive decisions (decisions that _p_). Second, rather than referring to algorithms as literally making decisions, ADM refers to the use of algorithms to solve decision problems at an organizational level. Third, since algorithmic tools by nature primarily settle cognitive decision problems, their classification as ADM depends on whether and to what extent an algorithmically generated output _p_ has an action triggering effect—i.e., translates into a consequential action _φ_. The examination of precisely this _p_-_φ_ relationship, allows us to pinpoint different ADM types (suggesting, offloading, superseding). Taking these three arguments into account, we arrive at the following definition: ADM refers to the practice of using algorithms to solve decision problems, where these algorithms can play a suggesting, offloading, or superseding role relative to humans, and decisions are defined as action triggering choices. (shrink)